blockjob.c

   1 /*
   2  * QEMU System Emulator block driver
   3  *
   4  * Copyright (c) 2011 IBM Corp.
   5  * Copyright (c) 2012 Red Hat, Inc.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "block/block.h"
  28 #include "block/blockjob_int.h"
  29 #include "block/block_int.h"
  30 #include "block/trace.h"
  31 #include "sysemu/block-backend.h"
  32 #include "qapi/error.h"
  33 #include "qapi/qapi-events-block-core.h"
  34 #include "qapi/qmp/qerror.h"
  35 #include "qemu/coroutine.h"
  36 #include "qemu/timer.h"
  37
  38 /*
  39  * The block job API is composed of two categories of functions.
  40  *
  41  * The first includes functions used by the monitor.  The monitor is
  42  * peculiar in that it accesses the block job list with block_job_get, and
  43  * therefore needs consistency across block_job_get and the actual operation
  44  * (e.g. block_job_set_speed).  The consistency is achieved with
  45  * aio_context_acquire/release.  These functions are declared in blockjob.h.
  46  *
  47  * The second includes functions used by the block job drivers and sometimes
  48  * by the core block layer.  These do not care about locking, because the
  49  * whole coroutine runs under the AioContext lock, and are declared in
  50  * blockjob_int.h.
  51  */
  52
  53 static bool is_block_job(Job *job)
  54 {
  55     return job_type(job) == JOB_TYPE_BACKUP ||
  56            job_type(job) == JOB_TYPE_COMMIT ||
  57            job_type(job) == JOB_TYPE_MIRROR ||
  58            job_type(job) == JOB_TYPE_STREAM;
  59 }
  60
  61 BlockJob *block_job_next(BlockJob *bjob)
  62 {
  63     Job *job = bjob ? &bjob->job : NULL;
  64
  65     do {
  66         job = job_next(job);
  67     } while (job && !is_block_job(job));
  68
  69     return job ? container_of(job, BlockJob, job) : NULL;
  70 }
  71
  72 BlockJob *block_job_get(const char *id)
  73 {
  74     Job *job = job_get(id);
  75
  76     if (job && is_block_job(job)) {
  77         return container_of(job, BlockJob, job);
  78     } else {
  79         return NULL;
  80     }
  81 }
  82
  83 void block_job_free(Job *job)
  84 {
  85     BlockJob *bjob = container_of(job, BlockJob, job);
  86
  87     block_job_remove_all_bdrv(bjob);
  88     blk_unref(bjob->blk);
  89     error_free(bjob->blocker);
  90 }
  91
  92 void block_job_drain(Job *job)
  93 {
  94     BlockJob *bjob = container_of(job, BlockJob, job);
  95     const JobDriver *drv = job->driver;
  96     BlockJobDriver *bjdrv = container_of(drv, BlockJobDriver, job_driver);
  97
  98     blk_drain(bjob->blk);
  99     if (bjdrv->drain) {
 100         bjdrv->drain(bjob);
 101     }
 102 }
 103
 104 static char *child_job_get_parent_desc(BdrvChild *c)
 105 {
 106     BlockJob *job = c->opaque;
 107     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
 108 }
 109
 110 static void child_job_drained_begin(BdrvChild *c)
 111 {
 112     BlockJob *job = c->opaque;
 113     job_pause(&job->job);
 114 }
 115
 116 static bool child_job_drained_poll(BdrvChild *c)
 117 {
 118     BlockJob *bjob = c->opaque;
 119     Job *job = &bjob->job;
 120     const BlockJobDriver *drv = block_job_driver(bjob);
 121
 122     /* An inactive or completed job doesn't have any pending requests. Jobs
 123      * with !job->busy are either already paused or have a pause point after
 124      * being reentered, so no job driver code will run before they pause. */
 125     if (!job->busy || job_is_completed(job)) {
 126         return false;
 127     }
 128
 129     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 130      * override this assumption. */
 131     if (drv->drained_poll) {
 132         return drv->drained_poll(bjob);
 133     } else {
 134         return true;
 135     }
 136 }
 137
 138 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
 139 {
 140     BlockJob *job = c->opaque;
 141     job_resume(&job->job);
 142 }
 143
 144 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 145                                       GSList **ignore, Error **errp)
 146 {
 147     BlockJob *job = c->opaque;
 148     GSList *l;
 149
 150     for (l = job->nodes; l; l = l->next) {
 151         BdrvChild *sibling = l->data;
 152         if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
 153             return false;
 154         }
 155     }
 156     return true;
 157 }
 158
 159 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 160                                   GSList **ignore)
 161 {
 162     BlockJob *job = c->opaque;
 163     GSList *l;
 164
 165     for (l = job->nodes; l; l = l->next) {
 166         BdrvChild *sibling = l->data;
 167         if (g_slist_find(*ignore, sibling)) {
 168             continue;
 169         }
 170         *ignore = g_slist_prepend(*ignore, sibling);
 171         bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
 172     }
 173
 174     job->job.aio_context = ctx;
 175 }
 176
 177 static const BdrvChildRole child_job = {
 178     .get_parent_desc    = child_job_get_parent_desc,
 179     .drained_begin      = child_job_drained_begin,
 180     .drained_poll       = child_job_drained_poll,
 181     .drained_end        = child_job_drained_end,
 182     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 183     .set_aio_ctx        = child_job_set_aio_ctx,
 184     .stay_at_node       = true,
 185 };
 186
 187 void block_job_remove_all_bdrv(BlockJob *job)
 188 {
 189     /*
 190      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 191      * which will also traverse job->nodes, so consume the list one by
 192      * one to make sure that such a concurrent access does not attempt
 193      * to process an already freed BdrvChild.
 194      */
 195     while (job->nodes) {
 196         GSList *l = job->nodes;
 197         BdrvChild *c = l->data;
 198
 199         job->nodes = l->next;
 200
 201         bdrv_op_unblock_all(c->bs, job->blocker);
 202         bdrv_root_unref_child(c);
 203
 204         g_slist_free_1(l);
 205     }
 206 }
 207
 208 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 209 {
 210     GSList *el;
 211
 212     for (el = job->nodes; el; el = el->next) {
 213         BdrvChild *c = el->data;
 214         if (c->bs == bs) {
 215             return true;
 216         }
 217     }
 218
 219     return false;
 220 }
 221
 222 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 223                        uint64_t perm, uint64_t shared_perm, Error **errp)
 224 {
 225     BdrvChild *c;
 226
 227     bdrv_ref(bs);
 228     if (job->job.aio_context != qemu_get_aio_context()) {
 229         aio_context_release(job->job.aio_context);
 230     }
 231     c = bdrv_root_attach_child(bs, name, &child_job, job->job.aio_context,
 232                                perm, shared_perm, job, errp);
 233     if (job->job.aio_context != qemu_get_aio_context()) {
 234         aio_context_acquire(job->job.aio_context);
 235     }
 236     if (c == NULL) {
 237         return -EPERM;
 238     }
 239
 240     job->nodes = g_slist_prepend(job->nodes, c);
 241     bdrv_op_block_all(bs, job->blocker);
 242
 243     return 0;
 244 }
 245
 246 static void block_job_on_idle(Notifier *n, void *opaque)
 247 {
 248     aio_wait_kick();
 249 }
 250
 251 bool block_job_is_internal(BlockJob *job)
 252 {
 253     return (job->job.id == NULL);
 254 }
 255
 256 const BlockJobDriver *block_job_driver(BlockJob *job)
 257 {
 258     return container_of(job->job.driver, BlockJobDriver, job_driver);
 259 }
 260
 261 /* Assumes the job_mutex is held */
 262 static bool job_timer_pending(Job *job)
 263 {
 264     return timer_pending(&job->sleep_timer);
 265 }
 266
 267 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 268 {
 269     int64_t old_speed = job->speed;
 270
 271     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 272         return;
 273     }
 274     if (speed < 0) {
 275         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 276         return;
 277     }
 278
 279     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 280
 281     job->speed = speed;
 282     if (speed && speed <= old_speed) {
 283         return;
 284     }
 285
 286     /* kick only if a timer is pending */
 287     job_enter_cond(&job->job, job_timer_pending);
 288 }
 289
 290 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 291 {
 292     if (!job->speed) {
 293         return 0;
 294     }
 295
 296     return ratelimit_calculate_delay(&job->limit, n);
 297 }
 298
 299 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 300 {
 301     BlockJobInfo *info;
 302
 303     if (block_job_is_internal(job)) {
 304         error_setg(errp, "Cannot query QEMU internal jobs");
 305         return NULL;
 306     }
 307     info = g_new0(BlockJobInfo, 1);
 308     info->type      = g_strdup(job_type_str(&job->job));
 309     info->device    = g_strdup(job->job.id);
 310     info->busy      = atomic_read(&job->job.busy);
 311     info->paused    = job->job.pause_count > 0;
 312     info->offset    = job->job.progress_current;
 313     info->len       = job->job.progress_total;
 314     info->speed     = job->speed;
 315     info->io_status = job->iostatus;
 316     info->ready     = job_is_ready(&job->job),
 317     info->status    = job->job.status;
 318     info->auto_finalize = job->job.auto_finalize;
 319     info->auto_dismiss  = job->job.auto_dismiss;
 320     info->has_error = job->job.ret != 0;
 321     info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 322     return info;
 323 }
 324
 325 static void block_job_iostatus_set_err(BlockJob *job, int error)
 326 {
 327     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 328         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 329                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 330     }
 331 }
 332
 333 static void block_job_event_cancelled(Notifier *n, void *opaque)
 334 {
 335     BlockJob *job = opaque;
 336
 337     if (block_job_is_internal(job)) {
 338         return;
 339     }
 340
 341     qapi_event_send_block_job_cancelled(job_type(&job->job),
 342                                         job->job.id,
 343                                         job->job.progress_total,
 344                                         job->job.progress_current,
 345                                         job->speed);
 346 }
 347
 348 static void block_job_event_completed(Notifier *n, void *opaque)
 349 {
 350     BlockJob *job = opaque;
 351     const char *msg = NULL;
 352
 353     if (block_job_is_internal(job)) {
 354         return;
 355     }
 356
 357     if (job->job.ret < 0) {
 358         msg = strerror(-job->job.ret);
 359     }
 360
 361     qapi_event_send_block_job_completed(job_type(&job->job),
 362                                         job->job.id,
 363                                         job->job.progress_total,
 364                                         job->job.progress_current,
 365                                         job->speed,
 366                                         !!msg,
 367                                         msg);
 368 }
 369
 370 static void block_job_event_pending(Notifier *n, void *opaque)
 371 {
 372     BlockJob *job = opaque;
 373
 374     if (block_job_is_internal(job)) {
 375         return;
 376     }
 377
 378     qapi_event_send_block_job_pending(job_type(&job->job),
 379                                       job->job.id);
 380 }
 381
 382 static void block_job_event_ready(Notifier *n, void *opaque)
 383 {
 384     BlockJob *job = opaque;
 385
 386     if (block_job_is_internal(job)) {
 387         return;
 388     }
 389
 390     qapi_event_send_block_job_ready(job_type(&job->job),
 391                                     job->job.id,
 392                                     job->job.progress_total,
 393                                     job->job.progress_current,
 394                                     job->speed);
 395 }
 396
 397
 398 /*
 399  * API for block job drivers and the block layer.  These functions are
 400  * declared in blockjob_int.h.
 401  */
 402
 403 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 404                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 405                        uint64_t shared_perm, int64_t speed, int flags,
 406                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 407 {
 408     BlockBackend *blk;
 409     BlockJob *job;
 410     int ret;
 411
 412     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 413         job_id = bdrv_get_device_name(bs);
 414     }
 415
 416     blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
 417     ret = blk_insert_bs(blk, bs, errp);
 418     if (ret < 0) {
 419         blk_unref(blk);
 420         return NULL;
 421     }
 422
 423     job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 424                      flags, cb, opaque, errp);
 425     if (job == NULL) {
 426         blk_unref(blk);
 427         return NULL;
 428     }
 429
 430     assert(is_block_job(&job->job));
 431     assert(job->job.driver->free == &block_job_free);
 432     assert(job->job.driver->user_resume == &block_job_user_resume);
 433     assert(job->job.driver->drain == &block_job_drain);
 434
 435     job->blk = blk;
 436
 437     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 438     job->finalize_completed_notifier.notify = block_job_event_completed;
 439     job->pending_notifier.notify = block_job_event_pending;
 440     job->ready_notifier.notify = block_job_event_ready;
 441     job->idle_notifier.notify = block_job_on_idle;
 442
 443     notifier_list_add(&job->job.on_finalize_cancelled,
 444                       &job->finalize_cancelled_notifier);
 445     notifier_list_add(&job->job.on_finalize_completed,
 446                       &job->finalize_completed_notifier);
 447     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 448     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 449     notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 450
 451     error_setg(&job->blocker, "block device is in use by block job: %s",
 452                job_type_str(&job->job));
 453     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 454
 455     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 456
 457     blk_set_allow_aio_context_change(blk, true);
 458
 459     /* Only set speed when necessary to avoid NotSupported error */
 460     if (speed != 0) {
 461         Error *local_err = NULL;
 462
 463         block_job_set_speed(job, speed, &local_err);
 464         if (local_err) {
 465             job_early_fail(&job->job);
 466             error_propagate(errp, local_err);
 467             return NULL;
 468         }
 469     }
 470
 471     return job;
 472 }
 473
 474 void block_job_iostatus_reset(BlockJob *job)
 475 {
 476     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 477         return;
 478     }
 479     assert(job->job.user_paused && job->job.pause_count > 0);
 480     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 481 }
 482
 483 void block_job_user_resume(Job *job)
 484 {
 485     BlockJob *bjob = container_of(job, BlockJob, job);
 486     block_job_iostatus_reset(bjob);
 487 }
 488
 489 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 490                                         int is_read, int error)
 491 {
 492     BlockErrorAction action;
 493
 494     switch (on_err) {
 495     case BLOCKDEV_ON_ERROR_ENOSPC:
 496     case BLOCKDEV_ON_ERROR_AUTO:
 497         action = (error == ENOSPC) ?
 498                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 499         break;
 500     case BLOCKDEV_ON_ERROR_STOP:
 501         action = BLOCK_ERROR_ACTION_STOP;
 502         break;
 503     case BLOCKDEV_ON_ERROR_REPORT:
 504         action = BLOCK_ERROR_ACTION_REPORT;
 505         break;
 506     case BLOCKDEV_ON_ERROR_IGNORE:
 507         action = BLOCK_ERROR_ACTION_IGNORE;
 508         break;
 509     default:
 510         abort();
 511     }
 512     if (!block_job_is_internal(job)) {
 513         qapi_event_send_block_job_error(job->job.id,
 514                                         is_read ? IO_OPERATION_TYPE_READ :
 515                                         IO_OPERATION_TYPE_WRITE,
 516                                         action);
 517     }
 518     if (action == BLOCK_ERROR_ACTION_STOP) {
 519         if (!job->job.user_paused) {
 520             job_pause(&job->job);
 521             /* make the pause user visible, which will be resumed from QMP. */
 522             job->job.user_paused = true;
 523         }
 524         block_job_iostatus_set_err(job, error);
 525     }
 526     return action;
 527 }